Why bacteria run Linux while eukaryotes run Windows

Why bacteria run Linux while eukaryotes run Windows? Sergei Maslov Brookhaven National Laboratory New York

Physical vs. Biological Laws Physical Laws are often discovered by finding simple common explanation for very different phenomena n Newton’s Law: n n Apples fall to the ground Planets revolve around the Sun Discovery of Biological Laws is slowed down by us having cookie-cutter explanation in terms of natural selection: 2

Drawing from Facebook group: Trust me, I'm a "Biologist"'

~ Genes encoded in bacterial genomes Packages installed on Linux computers 4

n n Complex systems have many components n Genes (Bacteria) n Software packages (Linux OS) Components do not work alone: they need to be assembled to work In individual systems only a subset of components is installed n Genome (Bacteria) – collection of genes n Computer (Linux OS) – collection of software packages Components have vastly different frequencies of installation 5

IKEA kits have many components Justin Pollard, http: //www. designboom. com 6

They need to be assembled to work Justin Pollard, http: //www. designboom. com 7

Different frequencies of use vs Common Rare 8

What determines the frequency of installation/use of a gene/package? n Popularity: AKA preferential attachment n n n Frequency ~ self-amplifying popularity Relevant for social systems: WWW links, facebook friendships, scientific citations Functional role: n n Frequency ~ breadth or importance of the functional role Relevant for biological and technological systems where selection adjusts undeserved popularity 9

Empirical data on component frequencies n Bacterial genomes (eggnog. embl. de): n n n Linux packages (popcon. ubuntu. com): n n n 500 sequenced prokaryotic genomes 44, 000 Orthologous Gene families 200, 000 Linux packages installed on 2, 000 individual computers Binary tables: component is either present or not in a given system 10

Frequency distributions Cloud Shell Core ORFans P(f)~ f-1. 5 except the top √N “universal” components with f~1 TY Pang, S. Maslov, PNAS (2013) 11

How to quantify functional importance? n n n We want to check Frequency ~ Importance Usefulness=Importance ~ Component is needed for proper functioning of other components Dependency network n n A B means A depends on B for its function Formalized for Linux software packages For metabolic enzymes given by upstreamdownstream positions in pathways Frequency ~ dependency degree, Kdep n Kdep = the total number of components that directly or indirectly depend on the selected one 12

TY Pang, S. Maslov, PNAS (2013) 13

Frequency is positively correlated with functional importance Correlation coefficient ~0. 4 for both Linux and genes Could be improved by using weighted dependency degree TY Pang, S. Maslov, PNAS (2013) 14

Warm-up: tree-like metabolic network TCA cycle Kdep=15 Kdep=5 TY Pang, S. Maslov, PNAS (2013) 15

Dependency degree distribution on a critical branching tree n P(K)~K-1. 5 for a critical branching tree n Paradox: Kmax-0. 5 ~ 1/N Kmax=N 2>N n Answer: parent tree size imposes a cutoff: there will be √N “core” nodes with Kmax=N n n present in almost all systems (ribosomal genes or core metabolic enzymes) Need a new model: in a tree D=1, while in real systems D~2>1 16

Bottom-down model of dependency network evolution n Components added gradually over evolutionary time New component directly depends on D previously existing components selected randomly Versions: n n n D is drawn from some distribution same as above Recent components are preferentially selected citations There is a fixed probability to connect to any previously existing components food webs 17

• p(t, T) –probability that component added at time T directly or indirectly depends on one added at time t 18

19

Kdep and Kout degree distributions 20

Kdep decreases layer number Linux Model with D=2 TY Pang, S. Maslov, PNAS (2013) 21

Zipf plot for Kdep distributions Metabolic enzymes vs Model Linux vs Model TY Pang, S. Maslov, PNAS (2013) 22

Frequency distributions Core Shell Cloud ORFans P(f)~ f-1. 5 except the top √N “universal” components with f~1 TY Pang, S. Maslov, PNAS (2013) 23

What experiments does P(f) help to interpret? 24

Pan-genome of E. coli strains M Touchon et al. PLo. S Genetics (2009)

Metagenomes The Human Microbiome Project Consortium, Nature (2012) 26

Pan-genome scaling 27

Pan-genome of all bacteria (# of genes in pan-genome) (# of new genes added to pan-genome) ~ (# of sequenced genomes)0. 5 ~ (# of sequenced genomes)-0. 5 P. Lapierre JP Gogarten TIG 2009 Slope=-0. 4 predictions of the toolbox model (-0. 5) 28

Bacterial genome evolution happens in cooperation with phages + =

Comparative genomics of E. coli implicates phages for Bit. Torrent 1 kb: gene length K-12 to B comparison Phage capacity: 20 kb Other strains up to 40 kb

WWW from AT&T website circa 1996 visualized by Mark Newman Phage-Bacteria Infection Network Data from Flores et al 2011 experiments by Moebus, Nattkemper, 1981

Why eukaryotes run windows? n Dependency network = reuse of components n n n Bacteria do not keep redundant genes after HGT Linux developers rely on previous efforts Pros: smaller genomes, open source, economies of scale Cons: less specialized, potentially unstable, “dependency hell” Eukaryotes are like Windows or Mac OS X n n Keep redundant components Proprietary software 32

# of pathways (or their regulators) Figure adapted from S. Maslov, TY Pang, K. Sneppen, S. Krishna, PNAS (2009) # of genes 33

Software packages for Linux Nselected packages ~ Ninstalled packages 1. 7 34

Collaborators: Tin Yau Pang, Stony Brook University Support: 35 Office of Biological and Environmental Research

Thank you!